At 8:30 AM on March 25, 2025, Unlimited Research Group of AI (URA) was honored to host a distinguished delegation of professors from the University of Bordeaux, France. The session featured three insightful talks that spanned a range of advanced research topics.
In the first talk, one of the professors delved into constructing detailed images using UNET models applied to MRI data. The discussion focused on the latest segmentation techniques, shedding light on cutting-edge methods currently being explored in the field.
The second talk centered on image classification using innovative AI techniques. The professor explained how regularization-based methods – by incorporating a penalty term into the loss function to keep the model close to its previous versions – help mitigate catastrophic forgetting. The talk also covered various approaches, including parameter and functional regularization, rehearsal/memory-based methods such as memory buffers and generative buffers, as well as architecture-based methodologies like ExpertGate, online continual learning, and the emerging move-to-data strategy, which collectively represent some of the state-of-the-art (SOTA) techniques under active investigation.
The third talk was delivered by Dr. Le Trong Nhan, head of ACLAB at HCMUT, who shared detailed insights into several IoT projects his laboratory has successfully implemented. His presentation highlighted the practical applications and future potential of IoT technologies in various domains.
The event concluded with an impressive live demonstration of URA’s virtual try-on product. This demo showcased a product that offered significantly higher accuracy and a more natural appearance compared to current SOTA models, with particular note of its performance when presenting challenging poses such as crossed arms.